AVRO-4269: [Java] Fix timestamp-nanos conversion before epoch

[emarussoo]

What is the purpose of the change

Fixes TimestampNanosConversion.toLong for Instant values before the Unix epoch with a positive nanosecond component.
The affected code path handles cases where seconds < 0 and nanos > 0. In that branch, the implementation needs to adjust the positive nanosecond component relative to the previous epoch second.
The previous implementation subtracted 1_000_000, which is a microsecond-scale constant. Since this conversion handles the timestamp-nanos logical type, the adjustment must subtract 1_000_000_000L, the number of nanoseconds in one second.
For example, 1969-12-31T23:59:59.999999999Z is one nanosecond before the Unix epoch and should convert to -1L. Before this fix, it converted to 998999999L.
This PR updates the adjustment constant and adds a regression test for the affected pre-epoch nanos case.
Fixes: AVRO-4269

Verifying this change

This change adds a regression test for timestamp-nanos conversion before the epoch.

This change added tests and can be verified as follows:

• mvn -Dtest=TestTimeConversions#timestampNanosConversionBeforeEpochWithPositiveNanos test
• mvn -Dtest=TestTimeConversions test

Documentation

• Does this pull request introduce a new feature? no
• If yes, how is the feature documented? not applicable

[emarussoo]

Hi everyone! No rush on the review, but whenever someone has a spare moment, could you please approve the workflows for this PR? Since I am a new contributor, the CI workflows need manual approval to run. It fixed the major issue reported in AVRO-4269. Thanks for your time!

[emarussoo]

Hi @nandorKollar, thank you for taking the time to review my contributions for AVRO-4269. Please let me know if there are any further adjustments or actions required from my side to proceed with the merge. Thanks again for your time!

[nandorKollar]

Merged to master, thanks @emarussoo for the fix.

[RyanSkraba]

Cherry-picked to branch-1.12.